应用LSTM和集成学习预测无创机械通气的吸呼气触发

Application of LSTM and Ensemble Learning to Predict Inspiratory and Expiratory Triggers for Noninvasive Mechanical Ventilation

机械通气是一种常用于辅助患者呼吸的治疗手段,但运行与患者呼吸异步时,会导致患者肺损伤等问题。本研究利用气体流量数据预测吸呼气触发,实现自适应地根据患者使用状态进行呼吸同步,并比较分析了长短期记忆网络(Long Short-TermMemory,LSTM)模型和不同集成学习模型的性能。结果表明,随机森林相较于其他研究的方法性能更好,模型预测的加权F1分数达到了0.98。

Mechanical ventilation is a treatment commonly used to assist patients’ breathing,and depending on the ventilation settings,it can lead to problems such as lung injury when it operates asynchronously with the patient’s breathing.In this study,we use gas flow timing data to predict inspiratory and expiratory triggers to achieve adaptive synchronization of breathing according to the patient’s state of use without additional adjustment of ventilation settings,and compare and analyze the performance of Long Short-Term Memory(LSTM) network models as well as different integrated learning models.The results show that although the LSTM neural network model has an advantage in its ability to face noisy interference,the random forest performs better compared to the other studied methods,and the model predicts a weighted F1 score of 0.98 on the simulated generated dataset with noisy interference.